The University of Bologna (UniBo) Deep Learning (DL) project. The food recognition challenge involves the building of a model to recognise and segment food items in each image. There are 273 categories of food items and each image may be a binary or multiclass image. More information on the challenge can be found on the official website.
The dataset used can be downloaded using the following link and consists of:
- train-v0.4.tar.gz : 24119 RGB images, with their corresponding 39328 annotations in MS-COCO format.
- val-v0.4.tar.gz : 1269 RGB images, with their corresponding 2053 annotations in MS-COCO format.
The steps taken are described in detail in the Report.
The code may also be viewed directly from the Notebook.
Example images below show the original image, the true mask, and the prediction by the trained model.
